Blood loss risk assessment method and system based on image data analysis

Abstract

The invention discloses a blood loss risk assessment method and system based on image data analysis, and relates to the technical field of image processing, and the method comprises the steps: recognizing and positioning the pixel coordinates of all checkerboard inner corners in a calibration board image and the corresponding world coordinates, solving a homography matrix, defining a color correction function, and carrying out the color correction. Forming a reference range interval of the healthy skin; based on a healthy skin reference range, judging a scalar field of each pixel position in the image, generating an initial wound area binary mask, and calculating a normalized chromaticity state field; and performing risk assessment on the bleeding state of the patient. According to the method, the problems of geometric distortion and color errors of the image are effectively solved by combining homography matrix correction and a color correction function, the precision of wound image analysis is improved, and the accuracy and stability of an evaluation result are enhanced by combining the local entropy yield and the diffusion flux vector field.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a blood loss risk assessment method and system based on image data analysis. BACKGROUND

[0002] With the rapid development of medical image analysis technology, disease diagnosis and risk assessment based on image data have gradually become an important means of clinical decision support, especially in the fields of wound monitoring and blood loss risk assessment. By using image processing and analysis technology, real-time and accurate disease state assessment can be provided for medical personnel. Computer vision technology combined with advanced algorithms such as deep learning, image segmentation and feature extraction has been widely used in automatic analysis of medical images.

[0003] The existing image-based blood loss risk assessment method still has deficiencies in actual application. Many methods fail to fully consider the geometric distortion and color error in the image shooting process when processing wound images, which limits the accuracy of the analysis results. Existing color correction methods often ignore the diversity of skin color in complex environments, making it difficult to accurately extract the subtle color difference between healthy skin and wound areas. SUMMARY

[0004] In view of the problems existing in the prior art, the present application arises at the historic moment.

[0005] Therefore, the present application provides a blood loss risk assessment method and system based on image data analysis to solve the problem that many methods fail to fully consider the geometric distortion and color error in the image shooting process when processing wound images, which limits the accuracy of the analysis results. Existing color correction methods often ignore the diversity of skin color in complex environments, making it difficult to accurately extract the subtle color difference between healthy skin and wound areas.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a blood loss risk assessment method based on image data analysis, comprising the following steps: Collecting patient images and performing preprocessing; The patient images include calibration plate images, 24-color card images, wound monitoring images and healthy skin RGB images; Identify and locate the pixel coordinates and corresponding world coordinates of all the inner corner points of the chessboard in the calibration plate image, solve the homography matrix, define the color correction function based on the 24-color card image, and process the healthy skin RGB image using the homography matrix and color correction function to determine the reference range interval of the healthy skin; The wound monitoring image is perspective transformed using a homography matrix, the geometrically corrected image is color corrected using a color correction function, a scalar field of each pixel position in the color corrected image is judged based on a healthy skin reference range, an initial wound area binary mask is generated, a wound area is constituted, and a normalized chroma state field is calculated; In the wound area pixel set, based on the normalized chroma state field, an original local entropy production rate is calculated, converted into a non-negative entropy production rate, the non-negative entropy production rate is taken as an initial state of a smoothing field, a diffusion flux vector field is calculated, and the smoothing field is updated based on the diffusion flux vector field to obtain a final local entropy production rate field.

[0007] As a preferred scheme of the blood loss risk assessment method based on image data analysis, wherein: the pixel coordinates and corresponding world coordinates of all the inner corner points of the checkerboard in the calibration plate image are identified and located, the homography matrix is solved, the color correction function is defined based on the 24-color card image, and the RGB image of the healthy skin is processed using the homography matrix and the color correction function to determine the reference range interval of the healthy skin, comprising: The Harris corner detection algorithm is applied to the calibration plate image to identify and locate the pixel coordinates and corresponding world coordinates of all the inner corner points of the checkerboard in the calibration plate image, and the homography matrix is solved based on the pixel coordinates and corresponding world coordinates; The 24-color card image is extracted by an image segmentation algorithm, the average value of all the RGB values in the color block region is calculated, the camera response vector of the color block is obtained, the corresponding standard Color space reference value; Based on the camera response vector, a corresponding second-order polynomial feature vector is constructed, the second-order polynomial feature vector is mapped to the standard Color space, the polynomial coefficient vector is calculated, and the color correction function is defined; The RGB image of the healthy skin is geometrically corrected using the homography matrix, the color of the geometrically corrected image is corrected using the color correction function, and the RGB value of each pixel of the image is converted to the standard Color space, and the skin image is obtained; In the skin image, the healthy skin region is demarcated using an artificial interactive ROI demarcation method as a reference region; The values of all the pixels in the reference region in the Channel are extracted to form a value set, the minimum value and the maximum value of the value set are screened to form a reference range interval of the healthy skin Channel value.

[0008] As a preferred embodiment of the blood loss risk assessment method based on image data analysis described in this invention, the following steps are included: performing perspective transformation on the wound monitoring image using a homography matrix; performing color correction on the geometrically corrected image using a color correction function; judging the scalar field of each pixel position in the color-corrected image based on a healthy skin reference range; generating an initial binary mask for the wound region; extracting pixel coordinates with a value of 1 to form the wound region; performing linear normalization; and calculating the normalized chromaticity state field, including: Perspective transformation was performed on the wound monitoring images using a homography matrix to obtain geometrically corrected images. Color correction was then applied to the geometrically corrected images using a color correction function to obtain... Color space image; from Extracting from color space images The channel components are used to obtain the scalar field. Based on the healthy skin reference range, the scalar field at each pixel position in the image is judged to generate an initial binary mask for the wound area. Morphological image processing is performed on the initial binary mask of the wound region to obtain the final binary mask of the wound region; Extract the coordinates of all pixels with a value of 1 from the final binary mask of the wound region to form the wound region; All pixels within the wound area The values ​​are linearly normalized to calculate the normalized chromaticity state field.

[0009] As a preferred embodiment of the blood loss risk assessment method based on image data analysis described in this invention, the following steps are included: Within the pixel set of the wound region, based on a normalized chromaticity state field, the original local entropy yield is calculated, converted into a non-negative entropy yield, and used as the initial state of the field to be smoothed. A diffusion flux vector field is then calculated, and the smoothing field is updated based on the diffusion flux vector field to obtain the final local entropy yield field. Within the pixel set of the wound region, the spatial gradient of the normalized chromaticity state field is approximated using the central difference method. Based on the principle of non-equilibrium thermodynamics, the negative spatial gradient of the normalized chromaticity state field is defined as the generalized thermodynamic force covariant vector of the current frame. The outer product of the spatial gradient of the normalized chromaticity state field and the transpose of the spatial gradient is obtained to obtain the outer product matrix. Gaussian smoothing is applied to the outer product matrix to calculate the local structure tensor. Regularize and invert the local structure tensor to obtain the Riemann metric tensor; Based on the normalized chromaticity state field, calculate the generalized thermodynamic flow inverse vector of the current frame; For each position, perform matrix and vector operations to calculate the original local entropy productivity. Apply the pointwise maximum function to each element of the original local entropy productivity to convert it into a non-negative entropy productivity. Using the non-negative entropy yield as the initial state of the field to be smoothed, the spatial gradient of the current iteration field is calculated using the central difference method. By combining the spatial gradient and Riemann metric tensor of the current iteration field, the diffusion flux vector field is calculated. Based on the explicit Euler scheme, the smoothing field is updated using the diffusion flux vector field to obtain the field for the next iteration. The number of iterations is set using a fixed iteration number method. When the number of iterations is reached Then stop iterating to obtain the final local entropy yield field.

[0010] As a preferred embodiment of the image data analysis-based blood loss risk assessment method of the present invention, wherein: defining the local entropy yield field as a thermodynamic risk map and generating a binary mask for the bleeding region includes: The local entropy yield field is defined as a thermodynamic risk map. All pixel values ​​located within the wound region in the thermodynamic risk map are extracted. The Otsu method is used to set the segmentation threshold to generate a binary mask for the bleeding region.

[0011] As a preferred embodiment of the image data analysis-based blood loss risk assessment method of the present invention, the step of assessing the patient's bleeding status includes: Extract the coordinates of all pixels with a value of 1 from the binary mask of the bleeding region to form the bleeding region; The sum of the local entropy production rates of all pixels within the bleeding area is calculated and defined as the blood loss risk index. Set alarm thresholds. Patients with a blood loss risk index greater than the alarm threshold are identified as having an abnormal bleeding state and an alarm signal is triggered. Patients with a blood loss risk index less than or equal to the alarm threshold are identified as having a normal state.

[0012] As a preferred embodiment of the blood loss risk assessment method based on image data analysis described in this invention, the step of acquiring and preprocessing patient images includes: Patient images were acquired using an RGB camera and then denoised and standardized.

[0013] Secondly, the present invention provides a blood loss risk assessment system based on image data analysis, comprising: The acquisition and preprocessing module is used to acquire patient images and perform noise reduction and standardization processing; The calibration reference area module is used for camera calibration using a checkerboard calibration board, calculating the homography matrix, performing geometric correction on the image, color correction using a color chart image, and converting to a standard. Color space, using an interactive ROI delineation method to extract healthy skin areas, and establish healthy skin Reference range for channel values; A detection state field module is configured to extract a wound area by image segmentation and morphological processing according to a healthy skin reference range, perform chroma normalization on the wound area, and calculate a chroma state field; A segmentation and local entropy production rate module is configured to calculate a generalized thermodynamic force covariant vector, calculate a local entropy production rate by a non-equilibrium thermodynamics model, and perform smoothing processing to obtain a final local entropy production rate field; A segmentation evaluation module is configured to generate a thermodynamic risk map according to the local entropy production rate field, perform image segmentation using the Otsu method, extract a bleeding area, calculate a sum of local entropy production rates of the bleeding area, generate a bleeding risk index, and perform risk evaluation on a bleeding state of a patient.

[0014] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the bleeding risk evaluation method based on image data analysis according to the first aspect of the present application.

[0015] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, and wherein the computer program, when executed by a processor, implements any step of the bleeding risk evaluation method based on image data analysis according to the first aspect of the present application.

[0016] The present application has the following beneficial effects: the present application effectively solves the problems of image geometric distortion and color error by combining homography matrix correction and color correction functions, improves the accuracy of wound image analysis, and enhances the accuracy and stability of the evaluation results by combining local entropy production rate and diffusion flux vector field. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0018] Figure 1 The figure is a running flowchart of the bleeding risk evaluation method based on image data analysis in embodiment 1.

[0019] Figure 2 The figure is a structure schematic diagram of the bleeding risk evaluation system based on image data analysis in embodiment 2. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0021] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods have not been described in detail in order not to unnecessarily obscure aspects of the present application.

[0022] Secondly, the "one embodiment" or "embodiment" referred to herein is intended to mean a specific feature, structure, or characteristic under discussion that is included in at least one implementation of the present application. The various appearances of "in one embodiment" or "in an embodiment" in the specification are not all referring to the same embodiment, although they can.

[0023] Embodiment 1, reference Figure 1 For the first embodiment of the present application, the embodiment provides a blood loss risk assessment method based on image data analysis, comprising the following steps: S1, collect patient images and pre-process, identify and locate the pixel coordinates and corresponding world coordinates of all the inner corner points of the checkerboard in the calibration board image, solve the homography matrix, define the color correction function based on the 24 color card image, and process the RGB image of healthy skin using the homography matrix and color correction function, determine the reference range interval of healthy skin; Specifically, collecting patient images and pre-processing includes: Use an RGB camera to collect patient images and perform denoising and standardization processing; The patient image includes a calibration board image, a 24 color card image, a wound monitoring image, and an RGB image of healthy skin; Place a standard 24 color card in the camera field of view and collect a 24 color card image; Place a planar checkerboard calibration board with a known physical size (e.g., each black and white square has a length of LrealL centimeters) in the camera field of view, place the checkerboard calibration board on a reference plane approximately parallel to the plane where the wound area is located, ensure that its surface is flat and completely within the imaging field of view, and collect the calibration board image.

[0024] Not only can high-quality patient images be obtained, but also through standardization and denoising preprocessing, the influence of external interference factors is reduced, ensuring the reliability and accuracy of image information in the subsequent processing process.

[0025] Further, identify and locate the pixel coordinates and corresponding world coordinates of all the inner corner points of the checkerboard in the calibration board image, solve the homography matrix, define the color correction function based on the 24 color card image, and process the RGB image of healthy skin using the homography matrix and color correction function, determine the reference range interval of healthy skin, including: The Harris corner detection algorithm is applied to the calibration board image to identify and locate the pixel coordinates and corresponding world coordinates of all the inner corner points in the checkerboard image. A world coordinate system is established with a corner of the checkerboard as the origin, and the plane is the calibration board plane. The projection relationship between the pixel coordinates and the world coordinates is described by a 3x3 homography matrix. Based on the pixel coordinates and the corresponding world coordinates, the homography matrix is solved, and the formula is: , where s is a non-zero homogeneous coordinate scale factor, and are the image pixel coordinates of the mth corner point, and are the world plane coordinates corresponding to the mth corner point, is the homography matrix, and the Levenberg-Marquardt nonlinear optimization algorithm is used to minimize the sum of the squared reprojection errors of all the corner points as the objective function for iterative optimization. The homography matrix is iteratively optimized, and the formula is: ,, where is the optimization objective function (sum of squared reprojection errors) of the homography matrix, M is the total number of checkerboard corner points, is the pixel coordinate of the mth detected checkerboard corner point, and m is the mth inner corner point of the checkerboard, is the world coordinate corresponding to the mth checkerboard corner point, is the reprojection pixel coordinate obtained by projecting the pixel coordinate of the checkerboard corner back to the image plane using the homography matrix; The 24 color block regions in the 24 color card image are extracted by the image segmentation algorithm, and the average value of all the RGB values in the color block region is calculated to obtain the camera response vector of the color block. The standard color space reference value corresponding to the color block is provided by the color card manufacturer, and the formula is: , , where is the camera response vector of the jth color block, , and are the average values of the red (R), green (G), and blue (B) channel values of all the pixels in the jth color block region, respectively, and T is the transpose operation. Based on the camera response vector, a corresponding second-order polynomial feature vector is constructed to establish a nonlinear mapping, and the formula is: , where is a constant term, R, G and B are three components of the camera response vector, representing the values of red, green and blue channels respectively, , and are the square terms of the channel values, , and are the cross terms between different channel values; Map the second order polynomial feature vector to the standard color space, calculate the polynomial coefficient vector, the formula is: , , , , , , wherein , and are the standard lightness value, the standard red-green component value and the standard yellow-blue component value respectively, j is the index of the color block on the standard color card, , and are the regression coefficient vectors of the channel, channel and channel, solved by least square method, is a general symbol, representing the coefficient vector used in polynomial regression, is the second order polynomial feature vector of the jth color block, is the second order polynomial feature mapping function; Define the color correction function, the formula is: , wherein is the color correction function; Use the homography matrix to perform geometric correction on the RGB image of healthy skin, use the color correction function to perform color correction on the geometrically corrected image, convert the RGB value of each pixel to the standard color space, obtain the skin image; In the skin image, use the artificial interactive ROI demarcation method to demarcate the healthy skin area as the reference area; Extract the values of all pixels in the reference area in the channel, form a value set, filter the minimum and maximum values of the value set, and form a healthy skin The reference range interval of the channel value is: wherein is the healthy skin The reference range interval of the channel value is and is the minimum value and the maximum value of the channel value of the healthy skin, respectively; The channel is a dimension in the standard color space, which represents the red-green opposite axis of the color, and the positive value represents the red direction and the negative value represents the green direction.

[0026] By accurately solving the homography matrix, the image error caused by the shooting angle and the field distortion is solved, and by combining the Harris corner point detection and the Levenberg-Marquardt nonlinear optimization algorithm, the accuracy of the corner point matching is greatly improved, thereby ensuring the efficiency and accuracy of the geometric correction, and through the threshold setting and the morphological processing in the image analysis, the accurate extraction of the wound area is ensured, and the interference of the noise and the artifacts is eliminated, which not only can monitor the change of the wound area in real time, but also can provide accurate blood loss risk assessment in a data-based manner to help medical staff discover bleeding abnormalities earlier.

[0027] S2, perspective transformation is performed on the wound monitoring image using the homography matrix, color correction is performed on the geometrically corrected image using a color correction function, judgment is made on the scalar field of each pixel position in the color corrected image based on the healthy skin reference range, an initial wound area binary mask is generated, pixel coordinates with a value of 1 are extracted to form a wound area, and linear normalization is performed to calculate a normalized chroma state field; Specifically, perspective transformation is performed on the wound monitoring image using the homography matrix, color correction is performed on the geometrically corrected image using a color correction function, judgment is made on the scalar field of each pixel position in the color corrected image based on the healthy skin reference range, an initial wound area binary mask is generated, pixel coordinates with a value of 1 are extracted to form a wound area, and linear normalization is performed to calculate a normalized chroma state field, including: The homography matrix is used to perform perspective transformation on the wound monitoring image to eliminate image distortion caused by the camera viewing angle and obtain a geometrically corrected image; The color correction function is used to perform color correction on the geometrically corrected image to convert the RGB value of each pixel to the standard color space to obtain a color space image; The color space image is extracted from the ​​Channel component, get scalar field, judge each pixel position in the image based on the healthy skin reference range, generate the initial wound area binary mask, the formula is: , Wherein is the initial wound area binary mask, indicating the binary indicator function (mask) on the image plane, the function value indicates whether the pixel is preliminarily judged to belong to the wound area at position , time , is the acquisition time corresponding to the k-th image, is the scalar field function on the image plane, and the function value is the channel value corresponding to the pixel of the image at position , time , is the two-dimensional pixel coordinate vector in the image plane, , x is the column index (horizontal direction) of the pixel, and y is the row index (vertical direction) of the pixel; Morphological image processing is performed on the initial wound area binary mask to eliminate noise, fill holes and smooth the region boundary to obtain the final wound area binary mask; All pixel coordinates with a value of 1 in the final wound area binary mask are extracted to form a wound area; The value of all pixels in the wound area is linearly normalized to map it to the interval , and the normalized chroma state field is calculated, which is the relative intensity of the overall chroma extreme value in the wound area in the current frame, and the formula is: , Wherein is the normalized chroma state field function on the wound area, and the value domain is strictly mapped to the interval , the value of 0 represents that the point has the most "dark" or least red chroma state in the wound area of the current frame (which may correspond to necrotic tissue or dark blood scab), and the value of 1 represents that the point has the most "red" chroma state in the wound area of the current frame (which may correspond to fresh bleeding), which eliminates the difference of absolute chroma value under different patients and different lighting conditions, focuses on the relative and dynamic chroma distribution structure in the wound, and provides a dimensionless and standardized state input for subsequent physical modeling based on gradient and diffusion, is a zero constant to prevent division by zero and avoid numerical calculation errors, and are the minimum and maximum values of the wound area, respectively.

[0028] The image distortion is eliminated by perspective transformation, so that the wound area can correspond to the actual physical space more accurately, eliminating the error that the visual angle may bring to the wound monitoring, and the color correction makes the color information in the image more stable and consistent, especially the color difference between the healthy skin and the wound area can be more accurately extracted. Through accurate calibration and processing, not only the healthy skin and the wound area can be accurately distinguished, but also the background noise and artifacts can be removed, ensuring the accurate extraction of the wound area. For long-term monitoring and dynamic change evaluation, this processing method makes the evaluation result more accurate, which can effectively reflect the healing process and change trend of the wound, and provide more accurate decision support for doctors.

[0029] S3、in the wound area pixel set, based on the normalized color state field, calculate the original local entropy production rate, convert to non-negative entropy production rate, take the non-negative entropy production rate as the initial state of the smoothing field, calculate the diffusion flux vector field, update the smoothing field based on the diffusion flux vector field, and obtain the final local entropy production rate field; Specifically, in the wound area pixel set, based on the normalized color state field, calculate the original local entropy production rate, convert to non-negative entropy production rate, take the non-negative entropy production rate as the initial state of the smoothing field, calculate the diffusion flux vector field, update the smoothing field based on the diffusion flux vector field, and obtain the final local entropy production rate field, including: In the wound area pixel set, the spatial gradient of the normalized color state field is approximated using the central difference method, and according to the principle of non-equilibrium thermodynamics, the negative spatial gradient of the normalized color state field is defined as the covariant vector of the generalized thermodynamic force of the current frame; The outer product of the spatial gradient of the normalized color state field and the transpose of the spatial gradient is obtained to obtain the outer product matrix, and the outer product matrix is Gaussian smoothed (filtered) to calculate the local structure tensor, and the formula is: , Wherein is the local structure tensor (2x2 matrix) at position and time , is a two-dimensional Gaussian kernel function with a standard deviation of , is a fixed parameter, is a convolution operator, is the spatial gradient of the normalized color state field; The local structure tensor is regularized and inverted to construct a Riemann metric tensor field describing the local geometry of the state space, and the Riemann metric tensor is obtained, and the formula is: , Wherein is the Riemann metric tensor at position and time Riemannian metric tensor at the location, Here, is the regularization parameter, a preset minimal positive constant, and I is a 2×2 identity matrix. Let I be the trace of the local structure tensor, which is the sum of its main diagonal elements, and let I be the identity matrix. This operation ensures It is always a positive definite and invertible matrix. The physical meaning of the metric g is: in the direction of drastic chromaticity change (active diffusion), A large eigenvalue of g implies a small eigenvalue of its inverse g, which means that the "thermodynamic generalized distance" or "dissipation resistance" in that direction is small. According to non-equilibrium thermodynamics, generalized flow is the time rate of change of state variables. Calculate the generalized thermodynamic flow inverse vector field of the current frame. Based on the normalized chromaticity state field, the generalized thermodynamic flux inverse vector of the current frame is calculated using the forward temporal difference method, as shown in the formula: , in For in position ,time The generalized thermodynamic flow inverse vector at that location. for Partial derivative with respect to time t; According to non-equilibrium thermodynamics, on a local Riemannian manifold defined by the metric g, the entropy production rate is given by the inner product of the generalized force X and the generalized flow Q induced by the metric. For each position Perform matrix and vector operations to calculate the original local entropy productivity, using the following formula: , in For in position ,time The original local entropy production rate at that location, For in position ,time The covariant component of the generalized thermodynamic force vector at a given point, where thermodynamic force is defined as the negative of the state gradient, i.e. ; Since the Riemann metric tensor g is a symmetric positive definite matrix, the above tensor operations are numerically equivalent to matrix-vector multiplication. The formula adopts the Einstein summation convention and sums over the repeating indices i and j. Applying a pointwise maximum function to each element of the original local entropy yield transforms it into a nonnegative entropy yield; Extract the original local entropy yield value, compare this value with the constant 0, and take the larger value in the comparison result as the final non-negative entropy yield value of the pixel position that conforms to the second law of thermodynamics; Using the non-negative entropy yield as the initial state of the field to be smoothed, the spatial gradient of the current iteration field is calculated using the central difference method. According to the anisotropic diffusion model in the form of Fick's law, the diffusion flux is related to the spatial gradient of the current iteration field through the Riemann metric tensor; Combining the spatial gradient and Riemann metric tensor of the current iteration field, the diffusion flux vector field is calculated using the following formula: , in For the diffusion flux vector field; According to the explicit Euler scheme, the smooth field is updated based on the diffusion flux vector field to obtain the field for the next iteration, as shown in the formula: , in For the (n+1)th iteration, at position Smooth field value at that point, This is the pseudo-time step, used to control the magnitude of each iteration update, where n is the number of iterations. The divergence (scalar field) of the diffusion flux vector field is calculated using the central difference method. For gradient operators; Use the fixed iteration number method to set the number of iterations. Using a rule of thumb, when the number of iterations is reached... Then stop iterating to obtain the final local entropy yield field.

[0030] By converting it into a nonnegative entropy yield, it not only conforms to the second law of thermodynamics but also avoids instability problems in numerical calculations. The calculation of the diffusion flux vector field can effectively capture the local thermodynamic changes in the wound region, and the accuracy of the evaluation results can be improved by updating the smooth field. Through multiple iterations and optimization calculations, the final result of the local entropy yield field will have higher stability and accuracy.

[0031] S4. Define the local entropy production field as a thermodynamic risk map, generate a binary mask of the bleeding area, and conduct risk assessment of the patient's bleeding status. Specifically, defining the local entropy production field as a thermodynamic risk map, the generation of a binary mask for the bleeding region includes: The local entropy yield field is defined as a thermodynamic risk map. All pixel values ​​within the wound region in the thermodynamic risk map are extracted. The Otsu method is used to set a segmentation threshold, generating a binary mask for the bleeding region. The formula is: , in For in position ,time Binary masking of the bleeding area. For a moment The wound area, For in position ,time Thermodynamic risk diagram at the location. To determine the segmentation threshold, all grayscale values ​​in the pixel values ​​within the wound area are used as candidate thresholds. All pixel values ​​within the wound area are divided into foreground and background parts. The inter-class variance of the two parts is calculated, and the candidate threshold that maximizes the inter-class variance is selected as the final threshold.

[0032] Through multiple iterations and optimization calculations, the final result of the local entropy production field will have higher stability and accuracy. Using Otsu's method for image segmentation has significant advantages because it can automatically determine the optimal threshold without human intervention. The blood loss risk index quantifies the thermodynamic changes in the bleeding area, thus providing a direct basis for assessing the patient's blood loss status.

[0033] Further risk assessment of the patient's bleeding status, including: Extract the coordinates of all pixels with a value of 1 from the binary mask of the bleeding region to form the bleeding region; The sum of the local entropy production rates of all pixels within the bleeding area is calculated and defined as the blood loss risk index. The statistical analysis method is used to calculate the mean of the historical blood loss risk index (starting from the time of monitoring, a data buffer is started to store each blood loss risk index received within 10 minutes in chronological order) plus twice the standard deviation. An alarm threshold is set. Patients whose blood loss risk index is greater than the alarm threshold are judged as having an abnormal bleeding state and an alarm signal is triggered. Patients whose blood loss risk index is less than or equal to the alarm threshold are judged as having a normal state.

[0034] By combining thermodynamic principles and image analysis, this invention can accurately capture microscopic changes in the wound area, providing a more scientific and objective assessment basis. Through the Otsu method to automatically set the segmentation threshold and the dynamic adjustment of the alarm threshold based on statistical methods, this invention greatly improves the automation level of the system and reduces the need for manual intervention. The threshold setting method based on standard deviation is highly adaptive and can adjust the alarm standard according to the specific situation of each patient, improving the flexibility and applicability of the system.

[0035] Example 2, refer to Figure 2 As a second embodiment of the present invention, a blood loss risk assessment system based on image data analysis includes: The acquisition and preprocessing module is used to acquire patient images and perform noise reduction and standardization processing; The calibration reference area module is used for camera calibration using a checkerboard calibration board, calculating the homography matrix, performing geometric correction on the image, color correction using a color chart image, and converting to a standard. Color space, using an interactive ROI delineation method to extract healthy skin areas, and establish healthy skin Reference range for channel values; The detection state field module is used to extract the wound area based on the healthy skin reference range through image segmentation and morphological processing, normalize the color of the wound area, and calculate the color state field. The segmentation and local entropy yield module is used to calculate the covariant vector of generalized thermodynamic forces, calculate the local entropy yield through a non-equilibrium thermodynamic model, and perform smoothing to obtain the final local entropy yield field. The segmentation assessment module is used to generate a thermodynamic risk map based on the local entropy production field, and to perform image segmentation using the Otsu method to extract the bleeding area, calculate the sum of the local entropy production of the bleeding area, generate a blood loss risk index, and assess the patient's bleeding status.

[0036] This embodiment also provides a computer device applicable to the blood loss risk assessment method based on image data analysis, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the blood loss risk assessment method based on image data analysis as proposed in the above embodiment.

[0037] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0038] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the blood loss risk assessment method based on image data analysis as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0039] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method of blood loss risk assessment based on image data analysis, characterized by: The method comprises the following steps: Collecting patient images and pre-processing; The patient images include calibration board images, 24-color card images, wound monitoring images, and RGB images of healthy skin; Identify and locate the pixel coordinates and corresponding world coordinates of all chessboard corners in the calibration board image, solve the homography matrix, define the color correction function based on the 24-color card image, and process the RGB image of healthy skin using the homography matrix and color correction function to determine the reference range interval of healthy skin; Perform perspective transformation on the wound monitoring image using the homography matrix, perform color correction on the geometrically corrected image using the color correction function, judge the scalar field of each pixel position in the color-corrected image based on the healthy skin reference range, generate an initial wound area binary mask, extract the pixel coordinates with a value of 1 to form the wound area, and perform linear normalization to calculate the normalized chroma state field; Within the wound area pixel set, based on the normalized chroma state field, calculate the original local entropy production rate, convert it to a non-negative entropy production rate, use the non-negative entropy production rate as the initial state of the smoothing field, calculate the diffusion flux vector field, update the smoothing field based on the diffusion flux vector field, and obtain the final local entropy production rate field; Define the local entropy production rate field as a thermodynamic risk map to generate a bleeding area binary mask and assess the bleeding state of the patient.

2. The image data analysis-based blood loss risk assessment method of claim 1, wherein: The identification and location of the pixel coordinates and corresponding world coordinates of all chessboard corners in the calibration board image, the solving of the homography matrix, the definition of the color correction function based on the 24-color card image, and the processing of the RGB image of healthy skin using the homography matrix and color correction function to determine the reference range interval of healthy skin, comprise: Apply the Harris corner detection algorithm to the calibration board image to identify and locate the pixel coordinates and corresponding world coordinates of all chessboard corners in the calibration board image, and solve the homography matrix based on the pixel coordinates and corresponding world coordinates; The 24 color block regions in the 24 color card image are extracted by an image segmentation algorithm, the average value of all RGB values in the color block region is calculated to obtain a camera response vector of the color block, and the standard color corresponding to the color block is provided by a color card manufacturer Color space reference value; Based on the camera response vector, a corresponding second order polynomial feature vector is constructed, the second order polynomial feature vector is mapped to a standard color space, a polynomial coefficient vector is calculated, and a color correction function is defined; The RGB image of healthy skin is geometrically corrected using a homography matrix, and the geometrically corrected image is color corrected using a color correction function to convert the RGB value of each pixel thereof to a standard color space to obtain a skin image; In the skin image, use the manual interactive ROI delineation method to delineate the healthy skin area as the reference area; Extract all pixels within the reference area The values ​​on the channel form a set of values. Filtering the set by finding the minimum and maximum values ​​results in healthy skin. Reference range for channel values.

3. The image data analysis-based blood loss risk assessment method of claim 2, wherein: The use of the homography matrix to perform perspective transformation on the wound monitoring image, the use of the color correction function to perform color correction on the geometrically corrected image, the judgment of the scalar field of each pixel position in the color-corrected image based on the healthy skin reference range, the generation of an initial wound area binary mask, the extraction of the pixel coordinates with a value of 1 to form the wound area, and the linear normalization to calculate the normalized chroma state field, comprise: The wound monitoring image is perspective transformed using a homography matrix to obtain a geometrically corrected image, and the geometrically corrected image is color corrected using a color correction function to obtain a color-corrected image color space image; From extracting channel components, obtaining a scalar field, judging the scalar field at each pixel position in the image based on a healthy skin reference range, and generating an initial wound region binary mask. Perform morphological image processing on the initial wound area binary mask to obtain the final wound area binary mask; Extract all pixel coordinates with a value of 1 in the final wound area binary mask to form the wound area; The normalized chrominance state field is calculated by linearly normalizing the values of all pixels within the wound area values of all pixels within the wound area.

4. The image data analysis-based blood loss risk assessment method of claim 3, wherein: The calculation of the original local entropy production rate within the wound area pixel set based on the normalized chroma state field, the conversion to a non-negative entropy production rate, the use of the non-negative entropy production rate as the initial state of the smoothing field, the calculation of the diffusion flux vector field, the updating of the smoothing field based on the diffusion flux vector field, and the obtaining of the final local entropy production rate field, comprise: Within the wound region pixel set, the spatial gradient of the normalized chromaticity state field is approximated using the central difference method, and according to the principle of non-equilibrium thermodynamics, the negative spatial gradient of the normalized chromaticity state field is defined as the generalized thermodynamic force covariant vector of the current frame; The spatial gradient of the normalized chromaticity state field is taken as the outer product with the transpose of the spatial gradient to obtain an outer product matrix, and the outer product matrix is Gaussian smoothed to calculate the local structure tensor; The local structure tensor is regularized and inverted to obtain the Riemann metric tensor; Based on the normalized chromaticity state field, the generalized thermodynamic flow inverse vector of the current frame is calculated; For each position, a matrix-vector operation is performed to calculate the original local entropy production rate, and a point-by-point maximum function is applied to each element of the original local entropy production rate to convert it to a non-negative entropy production rate; The non-negative entropy production rate is taken as the initial state of the field to be smoothed, and the spatial gradient of the current iteration field is calculated using the central difference method; The diffusion flux vector field is calculated in combination with the spatial gradient of the current iteration field and the Riemann metric tensor, the smoothed field is updated based on the diffusion flux vector field according to an explicit Euler format, a field of the next iteration is obtained, and the iteration number is set by using a fixed iteration number method When the iteration number is reached, the iteration is stopped, and a final local entropy production rate field is obtained.

5. The image data analysis-based blood loss risk assessment method of claim 4, wherein: The local entropy production rate field is defined as a thermodynamic risk map to generate a bleeding area binary mask, including: The local entropy production rate field is defined as a thermodynamic risk map, all pixel values located in the wound region in the thermodynamic risk map are extracted, a threshold is set using the Otsu method, and a bleeding area binary mask is generated.

6. The image data analysis-based blood loss risk assessment method of claim 5, wherein: The risk assessment of the bleeding state of the patient includes: All pixel coordinates with a value of 1 in the bleeding area binary mask are extracted to form a bleeding area; The sum of the local entropy production rates of all pixels in the bleeding area is calculated and defined as a blood loss risk index; An alarm threshold is set, patients with a blood loss risk index greater than the alarm threshold are determined to have an abnormal bleeding state, and an alarm signal is triggered, and patients with a blood loss risk index less than or equal to the alarm threshold are determined to be in a normal state.

7. The image data analysis-based blood loss risk assessment method of claim 1, wherein: The image of the patient is collected and preprocessed, including: An RGB camera is used to collect the image of the patient, and denoising and standardization processing are performed.

8. An image data analysis-based blood loss risk assessment system for implementing the image data analysis-based blood loss risk assessment method according to any one of claims 1 to 7, characterized by: It includes: A collection and preprocessing module is configured to collect the image of the patient and perform denoising and standardization processing; A correction reference region module is used for camera calibration using a checkerboard calibration board, calculating a homography matrix, geometric correction of an image, color correction through a color card image, conversion to a standard Color space, healthy skin area is extracted by manual interactive ROI demarcation method, and healthy skin Reference range of channel value; A state field detection module is configured to extract a wound region by image segmentation and morphological processing according to a healthy skin reference range, perform chromaticity normalization on the wound region, and calculate a chromaticity state field; A segmentation and local entropy production rate module is configured to calculate a generalized thermodynamic force covariant vector, calculate a local entropy production rate by a non-equilibrium thermodynamics model, and perform smoothing processing to obtain a final local entropy production rate field; A segmentation evaluation module is configured to generate a thermodynamic risk map based on the local entropy production rate field, perform image segmentation using the Otsu method, extract a bleeding area, calculate the sum of the local entropy production rates of the bleeding area, generate a blood loss risk index, and perform risk assessment of the bleeding state of the patient. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to implement the steps of the blood loss risk assessment method based on image data analysis of any one of claims 1 to 7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the blood loss risk assessment method based on image data analysis of any one of claims 1 to 7.

Images